A prediction method for the effect of enhancing coal seam permeability based on multi-source data fusion

Abstract

The present invention provides a method for predicting the coal seam permeability enhancement effect based on multi-source data fusion, belonging to the technical field of machine learning prediction. This method collects multi-source data of coal seam permeability enhancement in real time, normalizes the data, and then designs an encoding method for different types of data to extract key features. Then, the multi-source characterization data is uniformly serialized to align the time scales. The self-attention mechanism is used to calculate the time-dependent weights of single-modal data and dynamically adjust the importance of different data sources. Finally, the cross-attention mechanism is used to capture the information complementarity of cross-modal data. The fused data is input into the trained multi-parameter joint prediction model for coal seam permeability enhancement to output four key parameters, namely permeability, gas production, pressure change, and fracture density, for predicting and analyzing the coal seam permeability enhancement effect. The present invention realizes a comprehensive evaluation of the coal seam permeability enhancement effect. Compared with the existing methods, it can more accurately depict the coal seam permeability enhancement process and improve the prediction accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning prediction, and particularly relates to a prediction method for coal seam permeability enhancement effect based on multi-source data fusion. Background Technique

[0002] The accurate prediction of the coal seam permeability enhancement effect is of great significance for coalbed methane development and coal mine safety production. Coalbed methane, as a clean energy source, plays an important role in reducing greenhouse gas emissions and optimizing the energy structure. However, the coal seam itself has the characteristic of low permeability, and how to improve the permeability of the coal seam directly affects the extraction efficiency and economic benefits of coalbed methane. Therefore, scientifically and accurately predicting the coal seam permeability enhancement effect can not only guide the optimization of permeability enhancement transformation measures, but also improve the recovery rate of coalbed methane resources, reduce development costs, and enhance energy utilization efficiency. In addition, the prediction of the coal seam permeability enhancement effect is also crucial for coal mine safety production. During the coal mining process, the permeability enhancement transformation of the coal seam may cause changes in the geological structure, affect the stability of the coal seam, and even lead to safety hazards such as gas outbursts and rock collapses. Through effective prediction of the permeability enhancement effect, potential risks can be evaluated before construction, and a reasonable permeability enhancement plan can be formulated, thereby reducing the occurrence of safety accidents and ensuring the safety of coal mine workers. From the perspective of environmental protection, the efficient extraction of coalbed methane can reduce its disorderly emissions, reduce the impact of greenhouse gases on the atmospheric environment, and at the same time reduce the ecological damage caused by coal mine gas accidents. Therefore, the accurate prediction of the coal seam permeability enhancement effect not only helps to promote the efficient development of coalbed methane resources, but also promotes the coal mining industry to develop in a safer, greener and more sustainable direction.

[0003] At present, the prediction of coal seam permeability enhancement effect has become an important research direction in coalbed methane development and coal mine safety production. Traditional methods mainly rely on numerical simulation, experimental analysis and empirical formula calculation. Although these methods can evaluate the coal seam permeability enhancement effect to a certain extent, they often have problems such as complex calculation, poor adaptability and high dependence on on-site data, and it is difficult to meet the real-time prediction requirements under complex geological conditions. In addition, with the development of sensing technology, various types of sensors have been widely used in coal seam permeability monitoring, such as pressure sensors, temperature sensors, gas flow sensors, microseismic sensors and acoustic wave sensors, etc., providing rich data support for the prediction of coal seam permeability enhancement effect. However, the current research mainly focuses on the analysis of single or a small number of data sources, lacking the deep fusion and intelligent modeling of multi-source data, resulting in still great room for improvement in prediction accuracy and adaptability. In recent years, the progress of artificial intelligence technology has provided new ideas for the prediction of coal seam permeability enhancement effect. Methods based on machine learning and deep learning can automatically extract features from massive sensing data, improve prediction accuracy, and have a certain adaptive ability. However, such methods still face some challenges, such as the fusion modeling of multi-source heterogeneous data, abnormal data processing, missing data completion and multi-parameter joint prediction, etc. Therefore, how to effectively fuse various sensing data and construct a highly robust and accurate prediction model is still an important research direction at present. Summary of the Invention

[0004] In view of the above problems, the present invention proposes a method for predicting the coal seam permeability enhancement effect based on multi-source data fusion, including the following steps:

[0005] S1, Collect multi-source data of coal seam permeability enhancement in real time, including the pressure inside the pores and fractures of the coal seam , the coal seam temperature , the coal seam humidity , the coal seam gas flow , the vibration amplitude of the coal seam and the vibration frequency ;

[0006] S2, Normalize the data, then design corresponding coding methods for different types of data, extract key features, and construct a unified learnable representation;

[0007] S3, Serialize the multi-source representation data uniformly to align the time scales; then use the self-attention mechanism to calculate the time-dependent weights of single-modal data and dynamically adjust the importance of different data sources; finally, use the cross-attention mechanism to capture the information complementarity of cross-modal data;

[0008] S4. Input the fused data into the trained multi-parameter joint prediction model for coal seam permeability enhancement. The multi-parameter joint prediction model for coal seam permeability enhancement includes a multi-granularity time feature extraction network, a cross-attention mechanism layer, and a hierarchical linear regression layer. First, use the multi-granularity time feature extraction network to extract time-dependent features of different granularities from multi-source data. Secondly, use the cross-attention layer to achieve cross-fusion of features of different granularities. Finally, input the cross-fused features into the hierarchical linear regression layer to predict and output the predicted value of permeability , the predicted value of gas production , the predicted value of pressure change , the predicted value of fracture density Four key parameters;

[0009] S5. Based on the prediction output curves of the four key parameters in the time series, conduct prediction and analysis of the coal seam permeability enhancement effect.

[0010] Furthermore, the specific steps of S2 include:

[0011] First, encode the numerical type data; for numerical data ( ), design a learnable data segment Token with a dimension of 1024 dimensions, and realize the encoding of numerical type data by multiplying the Token and numerical data, mapping the numerical data to a high-dimensional space;

[0012] Then introduce the time series position encoding to capture the dynamic change characteristics of the time series; adopt the sine and cosine position encoding method to introduce explicit time information for different time steps. After combining the time series position encoding, obtain the final encoded representation;

[0013] Finally, use a multi-layer perceptron to perform mapping transformation on the encoded data, mapping it to a higher-dimensional feature expression. The MLP consists of two fully connected layers, and the activation function is ReLU; the obtained feature representation is the unified encoded feature of all data sources, serving as the basic representation for subsequent data fusion and input into the multi-parameter joint prediction model for coal seam permeability enhancement.

[0014] Furthermore, for the encoding of numerical type data, first perform regularization processing on the numerical data, and then perform learnable vector embedding Token multiplication encoding. Randomly initialize a 1024-dimensional learnable Token for different data types :

[0015] ;

[0016] Among them, is the high-dimensional feature after encoding.

[0017] Further, the specific process of S3 is as follows:

[0018] First, convert the multi-source characterization data into a unified time series, denoted as , where represents the data source, represents the time step the feature vector from the sensor at time , is the time series length, and the data of all sensors will be aligned according to the time step to form a multi-source sequence:

[0019] ;

[0020] Then, use the self-attention mechanism to calculate the fusion weights and the cross-attention mechanism for fusion: Finally, through a linear mapping layer, optimize the fused features to obtain the global correlation information of the multi-source data.

[0021] Further, an abnormal data determination and missing data supplementation process for multi-source data joint prediction is also included between S3 and S4; First, combine the encoded representations of the sensing data from other sources. Based on S3, determine whether a certain source of data is an outlier to ensure the reliability of the input data; Then, intelligently predict and complete the abnormally determined data and naturally missing data based on multi-source information.

[0022] Further, the specific process of the abnormal data determination for multi-source data joint prediction is as follows:

[0023] For the target variable to be determined, the historical time window of the current data data of time steps, and the historical data of the other 5 variables , obtain the corresponding representation based on S2 and S3;

[0024] Then, adopt the methods of time series modeling and multi-variable joint prediction to establish the correlation between the time series and the data, and construct a prediction model to predict the expected value of the target variable, and compare it with the observed value to determine whether there is an abnormality; The time series feature modeling uses Transformer to extract time series features and long-term dependence relationships, and constructs a multi-task joint prediction model through a multi-layer perceptron;

[0025] Calculate the prediction error , and then conduct error statistical analysis. The errors of normal data conform to the Gaussian distribution , and calculate the mean using historical data and the standard deviation ; After that, set the dynamic anomaly determination threshold:

[0026] ;

[0027] where is a hyperparameter that controls the degree of anomaly. When , it is determined as an outlier;

[0028] According to the degree of anomaly, set two anomaly levels: mild anomaly, that is, satisfying the condition ; Severe anomaly, that is, satisfying the condition ; If multiple variables are abnormal at the same time, it is necessary to analyze the global anomaly pattern to determine whether it is a systematic error.

[0029] Furthermore, the missing data filling process specifically includes:

[0030] Adopt a hybrid prediction method based on time dependence and spatial correlation to establish the correlation between time series, spatial relationship and data, and construct a spatio-temporal - multi-variable joint interpolation model ;

[0031] The spatio-temporal - multi-variable joint interpolation model includes a long short-term memory network and a recurrent neural network. The long short-term memory network is used to analyze the long-term time dependence relationship of different variables, and the recurrent neural network is used to analyze the spatial correlation of different variables to achieve high-precision interpolation of variable data;

[0032] At the same time, set a dynamic compensation mechanism. First, calculate the confidence. For each predicted value, calculate the corresponding prediction error;

[0033] If the determination of abnormal data in the previous joint prediction of multi-source data indicates that the prediction error of this variable data is lower than the dynamic anomaly determination threshold, then adjust its confidence during filling, and then perform multi-step prediction smoothing.

[0034] Furthermore, in the multi-parameter joint prediction model for enhancing coal seam permeability, the multi-granularity time feature extraction network includes a fully connected layer, a convolutional layer, a gated unit, and an LSTM layer; use the multi-granularity time feature extraction network to capture the short-term time dependence features contained in the fused time series representation , medium-term time dependence features and long-term time dependence features ; Among them, the fully connected layer is responsible for extracting primary features, and the convolutional layer, gated unit, and LSTM layer respectively perform deep feature extraction on the primary features to obtain , and ;

[0035] Then, the cross-attention mechanism layer is used to capture the correlation between multi-granularity features to obtain cross-fused features ;

[0036] The hierarchical linear regression layer designs four prediction branches with the same architecture but different parameters. Each branch corresponds to a target variable. For each target variable A hierarchical linear regression prediction branch is constructed, and its basic form is as follows:

[0037] ;

[0038] where is the input cross-fused feature, is the regression coefficient, is the regression bias, is the interval number index; is the number of intervals, indicating that different regression coefficients are used for fitting in different data ranges; each branch shares the underlying features but has an independent prediction layer; are the prediction results of four key parameters: permeability, gas production, pressure change, and fracture density;

[0039] When the multi-parameter joint prediction model for coal seam permeability enhancement is predicting, an outlier constraint and smoothing strategy are adopted. If the predicted value deviates from the historical trend, it will be corrected.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) Adaptive characterization construction of multi-type sensing data: Existing methods usually rely on manual feature engineering or single data source analysis, and it is difficult to effectively fuse multi-type data. The present invention designs different coding methods for different types of sensing data (such as numerical data, time series signals) to achieve automated and efficient data characterization construction. By means of numerical data normalization, time series data feature extraction, etc., the availability of data and the generalization ability of the model are improved;

[0042] (2) Adaptive interaction and fusion of multi-source data: Traditional methods often use simple weighting or splicing methods for multi-source data fusion, ignoring the dynamic relationship between different data sources. The present invention adopts the attention mechanism to perform sequential processing on multi-source data, so that the correlation between different sensor data can be fully mined. This method not only enhances the comprehensive expression ability of data, but also has stronger adaptability in complex environments;

[0043] (3)Abnormal data determination and missing data supplementation mechanism: The existing methods for dealing with abnormal data and missing data are relatively simple. For example, directly removing outliers or filling missing values with the mean may lead to information loss. The present invention utilizes the joint prediction mechanism of multi-source data to determine the outliers in a single data source and correct the anomalies through the information of other data sources. At the same time, for naturally missing data, the present invention uses a prediction model for intelligent completion to improve the integrity and accuracy of the data;

[0044] (4)Joint prediction of multiple parameters to improve prediction accuracy: Most traditional methods predict a single parameter, such as only predicting permeability or gas production, while ignoring the coupling relationship between multiple parameters. The present invention constructs a joint prediction model for multiple parameters, comprehensively considering four key parameters: permeability, gas production, pressure change, and fracture density, to achieve a comprehensive evaluation of the coal seam permeability enhancement effect. Compared with single-index prediction, this method provides richer information, helps to optimize the coal seam permeability enhancement plan, and improves the reliability and practical value of the prediction.

[0045] In summary, through innovative methods such as constructing adaptive representation, multi-source data interaction and fusion, intelligent determination and supplementation of abnormal data, and joint prediction of multiple parameters, the present invention solves the problems in the prior art such as difficult multi-source data fusion, difficult identification of abnormal data, and single prediction index, and realizes a more accurate, stable, and comprehensive prediction of the coal seam permeability enhancement effect, providing a reliable technical support for coalbed methane development and coal mine safety production. Description of the Drawings

[0046] Figure 1 It is the overall flowchart of the prediction method for the coal seam permeability enhancement effect based on multi-source data fusion.

[0047] Figure 2 It is the flowchart for collecting the multi-source data set for coal seam permeability enhancement.

[0048] Figure 3 It is the flowchart for encoding and representing multi-source data.

[0049] Figure 4 It is the flowchart for adaptive interaction and fusion of multi-source data.

[0050] Figure 5 It is the flowchart for determining abnormal data in the joint prediction of multi-source data.

[0051] Figure 6 It is the flowchart of the joint prediction model for multiple parameters of coal seam permeability enhancement.

[0052] Figure 7 It is the trend chart of the total loss change during the training process of the embodiment of the present invention.

[0053] Figure 8This is a graph showing the changing trend of the loss value of the test set during the training process of the embodiments of the present invention. Detailed implementation manners

[0054] The present invention will be further described below in conjunction with specific embodiments.

[0055] The present invention proposes a method for predicting the coal seam permeability enhancement effect based on multi-source data fusion. First, a multi-source data set is constructed by collecting and integrating data from pressure sensors, temperature sensors, gas flow sensors, microseismic sensors, and acoustic wave sensors. These data cover the key physical parameters during the coal seam permeability enhancement process, providing comprehensive input information for subsequent modeling. In the data representation stage, corresponding coding methods are designed for different types of data. For example, numerical data is normalized and mapped to a high-dimensional feature space, and time series signal data extracts time series features through convolutional or recurrent neural networks to achieve effective representation of different data types. In the data fusion stage, the present invention adopts an adaptive interactive fusion strategy, connects all sensing data into a unified sequence, and introduces an attention mechanism to mine the correlation between different data sources, realizing dynamic weighted fusion of cross-modal information. This method can enhance the expression ability of important features, while suppressing redundant or low-value information, and improving the adaptability of the model to complex permeability enhancement processes. For data anomalies and missing problems, the present invention proposes a joint prediction anomaly determination and data supplementation mechanism. By cross-comparing multi-source data, it is judged whether the data of a certain sensor deviates from the normal range, and the anomaly is further verified by combining time series pattern analysis. In addition, for the determined abnormal data and naturally missing data, an attention-based prediction model is used to complete them to ensure the integrity and reliability of the input data. Finally, the present invention constructs a multi-parameter joint prediction model, taking four key indicators of permeability, gas production, pressure change amount, and fracture density as target variables, comprehensively considering the mutual influence between parameters, and realizing accurate prediction of the coal seam permeability enhancement effect.

[0056] The overall process of this embodiment is as Figure 1 shown:

[0057] S1. Collection of multi-source data set for coal seam permeability enhancement; Deploy corresponding sensors to collect sensor data such as pressure, temperature, gas flow, microseismic, and acoustic wave as multi-source data, measure four indicators of permeability, gas production, pressure change amount, and fracture density as labels, and construct a multi-source data set;

[0058] S2. Coding and representation of multi-source data; Design corresponding coding methods for different types of data. Numerical data is first normalized, and key features are extracted from time series data to construct a unified learnable representation;

[0059] S3, Adaptive Interactive Fusion of Multi-source Data: First, uniformly serialize the multi-source data to align the time scales. Secondly, use the self-attention mechanism to calculate the time-dependent weights of single-modal data and dynamically adjust the importance of different data sources. Finally, use the cross-attention mechanism to capture the information complementarity of cross-modal data;

[0060] S4, Abnormal Data Judgment for Joint Prediction of Multi-source Data: Combine the encoded representations of sensing data from other sources. On the basis of S3, determine whether a certain source of data is an outlier to ensure the reliability of the input data;

[0061] S5, Missing Data Completion: Intelligently predict and complete the abnormal judgment data and natural missing data based on multi-source information to improve the integrity and consistency of the data;

[0062] S6, Multi-parameter Joint Prediction Model for Coal Seam Permeability Enhancement: This model includes a multi-granularity time feature extraction network, a cross-attention mechanism layer, and a hierarchical linear regression layer. First, use the multi-granularity time feature extraction network to extract the time-dependent features of different granularities of multi-source data. Secondly, use the cross-attention layer to achieve feature fusion of different granularities. Finally, input the fused features into the hierarchical linear regression layer to achieve multi-task prediction of four key parameters: permeability, gas production, pressure change, and fracture density;

[0063] S7, Model Deployment and Detection: Deploy the trained prediction model to the coal mine monitoring system to achieve real-time data input, automatic prediction, and abnormal warning.

[0064] The following specifically describes the specific implementation process of the present invention in combination with specific embodiments.

[0065] I. Collection of Multi-source Datasets for Coal Seam Permeability Enhancement:

[0066] 1. Deploy the following four types of sensors in the coal seam permeability enhancement operation area to monitor and collect key physical parameters:

[0067] A pressure sensor, used to measure the pressure change inside the coal seam pores and fractures, with the unit of MPa and a sampling frequency of 10 times per second (i.e., a sampling interval of 100 ms), denoted as ;

[0068] A temperature and humidity sensor, monitoring the coal seam temperature (unit: °C) and humidity (unit: %RH), with a sampling frequency of 10 times per second, denoted as , ;

[0069] A gas flow sensor, measuring the gas flow change in the coal seam and the air circulation condition, with the unit of m³ / h, used to evaluate the gas migration ability after coal seam permeability enhancement, with a sampling frequency of 10 times per second, denoted as ;

[0070] Microseismic sensors record microseismic events in coal seams, including vibration amplitude (unit: m / s²) and frequency (unit: Hz). The sampling frequency is 10 times per second, denoted as ;

[0071] 2. Data preprocessing: First, perform time series alignment. Since the sampling frequencies of different sensors may vary, timestamp synchronization is required to ensure data consistency;

[0072] 3. Data annotation: Combine the data from coal seam permeability enhancement experiments and actual manual measurements to annotate the samples, including four indicators: permeability, gas production, pressure change, and fracture density, denoted as respectively, in order to provide supervision signals for subsequent model training and data support for model evaluation;

[0073] The finally obtained data can be generally expressed as: where is the time step;

[0074] The complete process of collecting the entire multi-source dataset for coal seam permeability enhancement is as Figure 2 shown.

[0075] II. Encoding and characterization of multi-source data

[0076] 1. Encoding of numerical type data: For numerical data ( ), a learnable data segment Token with a dimension of 1024 is designed for each type of numerical data. The encoding of numerical type data is achieved by multiplying the Token with the numerical data, mapping the numerical data to a high-dimensional space;

[0077] First, perform numerical data regularization. Since the physical magnitudes of different data are different, first perform regularization processing on the numerical data:

[0078] ;

[0079] where represent the mean and standard deviation of the data respectively, ;

[0080] After that, perform learnable vector embedding Token multiplication encoding. Randomly initialize a 1024-dimensional learnable Token for different data types such as pressure, temperature, gas flow, etc. ;

[0081] ;

[0082] where is the The high-dimensional features of a data source is the Token embedding vector of the th data source;

[0083] 2. Encoding of time series data: Since the present invention performs a sampling operation on time series data at a frequency of 10 Hz, it is necessary to consider the temporal correlation between data, and obtain the pressure inside the coal seam pores and fractures , the coal seam temperature , the humidity of the coal seam , the change in the gas flow rate of the coal seam , the vibration amplitude of the coal seam , the vibration frequency of the coal seam . In addition to performing the above numerical encoding, it is also necessary to introduce temporal position encoding to capture the dynamic change characteristics of the time series. The present invention adopts a sine and cosine position encoding method to introduce explicit time information for different time steps:

[0084] ;

[0085] wherein, is the dimension of the encoded feature, is the time step, is the dimension index, is the sine position encoding, is the cosine position encoding;

[0086] Traverse the dimension index from 1 to 1024. When is odd, calculate the value of . When is even, calculate the value of . Arrange the calculated sine position encoding values and cosine position encoding values according to the dimension index to form a temporal position encoding of length 1024 ;

[0087] Combined with the temporal position encoding , the final encoded representation can be obtained as:

[0088] ;

[0089] wherein, is the time step, is the values of the previous time steps;

[0090] 3. Unified Coding Mapping: To uniformly map the coding features of different data sources and endow them with stronger expressive power, the present invention proposes to use an MLP (Multi-Layer Perceptron) to perform mapping transformation on the encoded data and map it to a feature representation in a higher dimension. The MLP consists of two fully connected layers, with the activation function being ReLU. The feature representation after mapping is:

[0091] ;

[0092] where and represent the weight and bias parameters of the first fully connected layer respectively, and represent the weight and bias parameters of the second fully connected layer respectively, is the intermediate coding feature of all data sources, is the th -step data of the th data source after temporal position encoding, randomly initialized, and the entire fully connected layer is learnable. The finally obtained feature representation

[0093] is the unified coding feature of all data sources and can be used as the basic representation for subsequent data fusion and model input; Figure 3 as shown in

[0094] III. Adaptive Interactive Fusion of Multi-Source Data

[0095] 1. Data Serialization: To enable unified processing of data from different sources, the present invention first converts it into a unified temporal sequence, denoted as , where represents the time step, represents the data source (sensor type), i.e., ; represents the feature vector of sensor at time step , is the temporal length. Finally, the data of all sensors will be aligned according to the time step to form a multi-source sequence:

[0096] ;

[0097] 2. Multi-source data feature fusion: Since the importance of different data sources may change over time, in order to dynamically capture the correlation between different data sources, the present invention uses the self-attention mechanism to calculate the fusion weights; in order to further enhance the information complementarity between different modal data, the present invention uses the cross-attention mechanism for fusion:

[0098] ;

[0099] Finally, through a linear mapping layer, the fused features are optimized:

[0100] ;

[0101] It contains the global correlation information of multi-source data, providing a unified and information-rich feature input for subsequent anomaly detection and prediction tasks;

[0102] The entire process of multi-source data adaptive interactive fusion is as Figure 4 shown.

[0103] IV. Determination of abnormal data in multi-source data joint prediction

[0104] This step aims to use the historical information of the current data and the joint information of other data sources to perform anomaly detection on various monitoring data inside the coal seam pores and fractures through time series modeling and multivariate relationship analysis, screen out unreasonable data points, and ensure the authenticity and reliability of the data.

[0105] Data input and feature extraction: For the target variable ( ), the input data includes the other 5 sensor data at the current time step and the historical time window of the current data time steps of data ; and the historical data of the other 5 variables . According to the steps of S2 and S3, the corresponding representations can be obtained for these input data.

[0106] Time series - multivariate joint anomaly detection model: In this step, a time series modeling + multivariate joint prediction method is adopted to establish the correlation between the time series and the data, and a prediction model is constructed to predict the expected value of the target variable and compare it with the observed value to determine whether there is an anomaly;

[0107] ;

[0108] That is, the predicted target variable not only depends on the other five variables, but also uses its own historical trend for prediction;

[0109] For time series feature modeling, the present invention uses a Transformer to extract time series features and long-term dependencies, and constructs a multi-task joint prediction model through a multi-layer perceptron:

[0110]

[0111] Abnormal data determination and processing: Calculate the prediction error , and then perform error statistical analysis. The errors of normal data conform to the Gaussian distribution , and use historical data to calculate the mean and the standard deviation . Then, set the dynamic abnormal determination threshold:

[0112] ;

[0113] where is a hyperparameter for controlling the degree of abnormality. In the present invention, is taken, which represents the 99.7% confidence interval. When , it is determined as an outlier;

[0114] According to the degree of abnormality, the present invention sets two abnormality levels: mild abnormality, that is, satisfying the condition ; severe abnormality, that is, satisfying the condition . If multiple variables are abnormal at the same time, it is necessary to analyze the global abnormal pattern to determine whether it is a systematic error. At the same time, the determined abnormal data is sent to the missing data supplement part and corrected by interpolation prediction;

[0115] This method combines the time trend and multi-variable relationship to improve the accuracy of abnormal detection and provide high-quality data support for the prediction of coal seam permeability enhancement effect. Its overall process is as Figure 5 shown.

[0116] V. Missing data supplement

[0117] This step aims to construct a spatio-temporal multi-variable joint interpolation model to supplement the missing data, ensure the integrity of the data, and provide high-quality data input for the subsequent joint prediction of permeability, gas production, pressure change amount, and fracture density. The missing data mainly includes: abnormal data correction (from the abnormal data detection result); natural missing data (data gaps caused by sensor failures, signal losses, etc.).

[0118] Space-time - Multivariate Joint Interpolation Model: In this step, a hybrid prediction method based on time dependence and spatial correlation is adopted to establish the correlation between time series, spatial relationships, and data, and a prediction model is constructed. ;

[0119] ;

[0120] That is, when predicting the target variable , it not only depends on the other 5 variables but also utilizes the time dependence and spatial correlation of its own sequence.

[0121] The space-time - multivariate joint interpolation model includes a long short-term memory network and a recurrent neural network. The long short-term memory network is used to analyze the long-term time dependence relationship of different variables, and the recurrent neural network is used to analyze the spatial correlation of different variables to achieve high-precision interpolation of variable data:

[0122] ;

[0123] Among them represents the long short-term memory network, represents the recurrent neural network.

[0124] Dynamic Uncertainty Compensation Mechanism: Since there is a certain degree of uncertainty in the predicted values of missing data, a dynamic compensation mechanism is set up to improve the reliability of the completed data. First, confidence calculation is carried out. For each predicted value , the corresponding prediction error is calculated:

[0125] ;

[0126] If the determination of abnormal data in the previous joint prediction of multi-source data indicates that the prediction error of this variable data is lower than the dynamic abnormal determination threshold, then its confidence is adjusted during completion:

[0127] ;

[0128] Among them, is the calculated dynamic abnormal threshold; is the confidence compensation factor (dynamically adjusted according to historical data), is the time series of the predicted value of the target variable, is the time series of the predicted value of the target variable after confidence adjustment;

[0129] After that, multi-step prediction smoothing is carried out. If the data is missing for a long time, single-step prediction may cause error accumulation, so multi-step prediction smoothing is adopted:

[0130] ;

[0131] Among them, For the smoothing window size, it is set to 5 steps in the present invention, that is .

[0132] VI. Multi-parameter joint prediction model for layer antireflection

[0133] Based on the processed multi-source data representation, this step constructs a multi-parameter joint prediction model for coal seam antireflection to realize the synchronous prediction of permeability , gas production , pressure change , fracture density . At the same time, a prediction result correction and smoothing strategy is introduced for the four prediction data reflecting the coal seam antireflection effect among the above four items. The multi-parameter joint prediction model for coal seam antireflection includes a multi-granularity time feature extraction network, a multi-head attention mechanism layer, and a hierarchical linear regression layer, combining time series modeling and multi-variable relationship modeling to accurately depict the dynamic change trend of the coal seam antireflection effect.

[0134] Time series feature extraction: After the input data is processed by S1-S5, the fused time series representation is obtained:

[0135] ;

[0136] Among them, is the fused feature vector at the current time , containing the time series information of multi-source data;

[0137] The multi-granularity time feature extraction network includes a fully connected layer, a convolutional layer, a gated unit, and an LSTM layer. The multi-granularity time feature extraction network is used to capture the short-term time-dependent features , medium-term time-dependent features and long-term time-dependent features contained in ; among them, the fully connected layer is responsible for extracting the primary features of , and the convolutional layer, gated unit, and LSTM layer respectively perform deep feature extraction on the primary features to obtain , specifically as follows:

[0138] ;

[0139] );

[0140] );

[0141] Among them represents the fully connected layer, represents the gated unit, represents the convolutional layer;

[0142] The cross-attention mechanism layer is used to capture the correlation between multi-granularity features to obtain cross features ;

[0143]

[0144] Piecewise linear regression layer: For the extracted cross features Four-task joint prediction is performed. Specifically, four prediction branches with the same architecture but different parameters are designed. Here, the piecewise linear regression method is adopted to improve the interpretability and stability of the prediction. Each branch corresponds to a target variable. For each target variable , where is the predicted value of permeability, is the predicted value of gas production, is the predicted value of pressure change, is the predicted value of fracture density. A piecewise linear regression prediction (Piecewise Linear Regression, PLR) branch is constructed, and its basic form is as follows:

[0145] ;

[0146] is the regression coefficient, is the bias of the regression, is the interval number index; is the prediction results of the four key parameters of permeability, gas production, pressure change, and fracture density; is the number of intervals, indicating that different regression coefficients are used for fitting in different data ranges. In the present invention, is taken; Each branch shares the underlying features but has an independent prediction layer to enhance the pertinence of the prediction and ensure the stability of the prediction.

[0147] Loss function design: Since the four target variables are correlated, a weighted loss of multi-task learning is adopted:

[0148] ;

[0149] Among them, the loss of each task adopts the mean square error (MSE). For the loss weights, the present invention adopts ;

[0150] ;

[0151] Among them, is the predicted value of the th data source at time, is the th data source at The true value at a moment.

[0152] Prediction result correction and output: To improve the stability and accuracy of prediction, an outlier constraint and smoothing strategy are adopted. If the predicted value significantly deviates from the historical trend, it is corrected:

[0153] ;

[0154] Among them, Controls the smoothing weight to ensure that the predicted value does not mutate. In the present invention, , is the th data source at the moment's true value.

[0155] For the short-term trend, a smoothing operation is also carried out, using a moving window average. This strategy can reduce high-frequency noise and improve the continuity of the prediction curve;

[0156] ;

[0157] Among them, is the th data source at the moment's predicted value. Finally, the prediction result is output to generate the prediction result at the future moment: , where is the permeability prediction time series at the future moment, is the gas production prediction time series at the future moment, is the pressure change prediction time series at the future moment, is the fracture density prediction time series at the future moment; The prediction results are used for the optimization decision of coal seam transformation, such as evaluating the permeability improvement trend, analyzing the change of coal seam gas production, judging the impact of pressure adjustment on coal seam permeability enhancement, predicting the evolution trend of fracture density, etc. Specifically:

[0158] Permeability improvement trend: Permeability is a key indicator for gas flow in coal seams and directly affects the gas extraction efficiency of coal seams. By predicting the permeability improvement trend, it can be understood whether the coal seam transformation effectively improves gas fluidity, thereby judging whether the coal seam permeability enhancement effect reaches the expected goal. If the permeability is low, it is necessary to further optimize the fracture structure of the coal seam or take other measures to increase gas channels;

[0159] Gas production: The gas production of coal seams is a direct manifestation of the enhanced permeability effect of coal seams. An increase in gas production usually means that the gas permeability and porosity of the coal seams have been improved. By analyzing the changes in gas production, the degree of enhanced permeability of the coal seams can be evaluated, and then decisions can be made on whether to adjust mining strategies, transformation plans, or pressure control strategies;

[0160] The impact of pressure change on enhanced coal seam permeability: The pressure change has a significant impact on the gas permeability and gas release capacity of coal seams. An appropriate pressure reduction can prompt the coal seam to release more gas, but an excessive pressure change may cause damage to the coal seam or unstable gas flow. By predicting the relationship between the pressure change and the enhanced permeability effect of the coal seam, decision-makers can reasonably adjust the fracturing pressure or gas injection pressure to optimize the enhanced permeability effect of the coal seam;

[0161] The evolution trend of fracture density: Fracture density is one of the physical properties of coal seams, directly affecting the gas permeability and connectivity of fracture channels in coal seams. Too low fracture density will make it difficult for coal seam gas to flow, affecting the enhanced permeability effect; while too high fracture density may lead to coal seam instability, even causing collapse or gas leakage. By predicting the evolution trend of fracture density, decision-makers can better evaluate the health status of fractures and take timely measures to repair or adjust the fracture structure of the coal seam to ensure the long-term effect of coal seam transformation.

[0162] Generally speaking, the combined prediction of these four parameters provides a comprehensive evaluation framework for the enhanced permeability effect of coal seams, helping decision-makers accurately adjust transformation strategies, optimize the gas permeability of coal seams, maximize the mining benefits of coal seam gas, and at the same time ensure the long-term stability and safety of coal seams.

[0163] Through the Transformer structure and multi-task learning method, the combined prediction of the four core parameters of enhanced coal seam permeability is realized, and the prediction results are optimized by anomaly detection and smoothing strategies to ensure the accuracy and stability of the prediction, providing data support for the optimization of enhanced coal seam permeability. The data processing flow chart of the combined multi-parameter prediction model for enhanced coal seam permeability is as Figure 6 shown.

[0164] Deploy the trained combined multi-parameter prediction model for enhanced coal seam permeability to the coal mine monitoring system to achieve real-time data input, automatic prediction, and anomaly warning. First, the model needs to be encapsulated and optimized to adapt to the actual industrial deployment environment. The model is converted into ONNX and TensorRT formats to improve the inference efficiency, and the prediction results are applied to real-time visualization and anomaly warning. When the predicted value deviates from the set threshold, an alarm is automatically triggered, and combined with the anomaly detection module using S4 - S5 multiplexing, false alarms are reduced and the detection accuracy is improved.

[0165] VII. Experimental results prove

[0166] This experiment is based on a multi-source sensor monitoring system actually deployed at a coal mine site. Over a period of 15 days, relevant data related to coal seam permeability enhancement was continuously collected, including six multi-source time-series data items such as pressure (P), temperature and humidity (T, W), gas flow rate (F), vibration amplitude (M), and vibration frequency (f), as well as four target parameters including permeability (K), gas production (Q), pressure change (ΔP), and fracture density (D). The sampling frequency of the data acquisition system was set at 10 Hz, and the ratio of training data to test data was 9:1.

[0167] During the training process, the change curve of the total loss value is as shown in Figure 7 As shown, correspondingly, after each round of training is completed, a test is conducted, and the change process of the corresponding test loss value is as shown in Figure 8 As shown. From Figure 7 and Figure 8 it can be seen that during the entire training process, the value of the loss function is continuously decreasing, the model is continuously converging, and finally it tends to be stable.

[0168] In terms of model performance evaluation, verification was carried out from three aspects: multi-parameter prediction, abnormal data determination, and missing data completion. For the multi-parameter prediction task, the following three error evaluation indicators were used for analysis: Mean Absolute Error (MAE): Measures the average absolute error between the predicted value and the true value. The smaller the value, the more accurate the prediction result; Mean Squared Error (MSE): Calculates the average of the squared errors between the predicted value and the true value, which can magnify the influence of larger errors on the overall error to evaluate the stability of the model; Root Mean Squared Error (RMSE): Is the square root of MSE, measures the overall level of prediction error, and keeps the unit consistent with the original data, more intuitively reflecting the prediction accuracy. The experimental results are shown in Table 1.

[0169] Table 1 Experimental results of multi-parameter prediction

[0170]

[0171] For the abnormal data determination task, Accuracy was used as the core evaluation indicator, and the experimental results are shown in Table 2.

[0172] Table 2 Experimental results of abnormal data determination accuracy

[0173]

[0174] For the missing data completion task, MAE and MSE are used to evaluate the completion effect of the model in the case of missing data. The experimental results are shown in Table 3. Lower MAE and MSE indicate that the data completed by the model is close to the true value, improving data integrity and ensuring the reliability of the prediction results.

[0175] Table 3 Experimental Results of Missing Data Completion

[0176]

[0177] The experimental results show that the method of the present invention achieves lower MAE, MSE, and RMSE values in the four core parameter prediction tasks, proving that the model can accurately predict the coal seam permeability enhancement effect; in the abnormal data determination task, the model has a high accuracy, effectively avoiding the interference of abnormal data; in the missing data completion task, the MAE and MSE errors are small, and the completed data is highly consistent with the real data. In summary, the experiment verifies the effectiveness and practical application value of the present invention in the coal seam permeability enhancement effect prediction task.

[0178] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0179] Although the specific implementation manners of the present invention are described above, they are not limitations on the protection scope of the present invention. Those skilled in the art should understand that, based on the technical solutions of the present invention, various modifications or deformations that can be made without creative efforts by those skilled in the art are still within the protection scope of the present invention.

Claims

1. A method for predicting coal seam permeability enhancement effect based on multi-source data fusion, characterized in that: The following steps are involved: S1, real-time collection of multi-source data on coal seam permeability enhancement, including pressure inside coal seam pores and fractures , coal seam temperature , Coal Seam Humidity , coal seam gas flow , Coal seam vibration amplitude and vibration frequency ; S2, normalize the data, then design corresponding encoding methods for different types of data, extract key features, and construct a unified learnable representation; First, encode the numerical data; for numerical data , for each type of data, a learnable data fragment Token with a dimension of 1024 is designed, and the encoding of the numerical type data is realized by multiplying the Token and the numerical data, and the numerical data is mapped to a high-dimensional space; Then, temporal position coding is introduced to capture the dynamic characteristics of time series. Sine and cosine position coding are used to introduce explicit time information for different time steps. After combining with temporal position coding, the final coding representation is obtained. Finally, a multi-layer perceptron is used to map the encoded data to a higher-dimensional feature expression. The MLP consists of two fully connected layers, and the activation function is ReLU; the obtained feature representation is It is the unified coding feature of all data sources, serving as the basic representation of the subsequent data fusion and the input of the multi-parameter joint prediction model of coal seam permeability enhancement; S3, uniformly serializes multi-source representation data and aligns time scales; then uses the self-attention mechanism to calculate the time-dependent weight of single-modal data and dynamically adjusts the importance of different data sources; finally, the cross-attention mechanism is used to capture the information complementarity of cross-modal data; S4, input the fused data into the trained coal seam permeability enhancement multi-parameter joint prediction model, which includes a multi-granularity time feature extraction network, a cross-attention mechanism layer and a hierarchical linear regression layer; firstly, a multi-granularity time feature extraction network is used to extract time-dependent features of different granularities of multi-source data; Secondly, the cross-attention layer is used to realize the cross-fusion of features of different granularities; finally, the cross-fusion features are input into the hierarchical linear regression layer to predict the output permeability prediction value , Gas production forecast , Prediction of pressure change , crack density prediction value Four key parameters; S5, based on the prediction output curves of the four key parameters of the time series, the coal seam permeability enhancement effect is predicted and analyzed.

2. The method for predicting coal seam permeability enhancement effect based on multi-source data fusion according to claim 1, characterized in that: The encoding of numerical data requires first regularizing the numerical data, and then performing learnable vector embedding Token multiplication encoding to encode the data of different data types. Randomly initialize a 1024-dimensional learnable Token : in, It is the high-dimensional feature after encoding.

3. The method for predicting coal seam permeability enhancement effect based on multi-source data fusion according to claim 1, characterized in that: The specific process of S3 is as follows: First, the multi-source representation data is converted into a unified time series, denoted as ,in Indicates the source of data. Represents the time step Moments from sensors The characteristic vector of is the time series length, and the data of all sensors will be in time steps Align to form a multi-source sequence: Then, the self-attention mechanism is used to calculate the fusion weights, and the cross-attention mechanism is used for fusion: finally, a linear mapping layer is used to optimize the fusion features and obtain the global correlation information of multi-source data.

4. A method for predicting coal seam permeability enhancement effect based on multi-source data fusion as claimed in claim 3, characterized in that: The process between S3 and S4 also includes abnormal data determination and missing data supplementation for joint prediction of multi-source data. First, based on S3, the sensor data encoding representation of other sources is combined to determine whether a certain source data is an abnormal value to ensure the reliability of the input data. Then, based on multi-source information, intelligent prediction and completion are performed on abnormal judgment data and naturally missing data.

5. The method for predicting coal seam permeability enhancement effect based on multi-source data fusion according to claim 4, characterized in that: The specific process of abnormal data determination in the multi-source data joint prediction is as follows: For the target variable to be determined , the historical time window of the current data time step data And the historical data of the remaining 5 variables , based on S2 and S3, the corresponding characterization is obtained ; Then, we use time series modeling and multivariate joint prediction to establish the correlation between time series and data and build a prediction model. , predict the expected value of the target variable , and the observed value Compare and determine whether there is an abnormality; the time series modeling uses Transformer to extract time series features and long-term dependencies, and builds a multi-task joint prediction model through a multi-layer perceptron; Calculating prediction error , then the error statistics analysis, the error of normal data Gaussian distribution , using historical data to calculate the mean and standard deviation ; Then, set the dynamic anomaly determination threshold: in is a hyperparameter that controls the degree of abnormality. When , it is judged as an abnormal value; According to the degree of abnormality, two abnormality levels are set: mild abnormality, that is, meeting the conditions ; Severe abnormality, that is, meeting the conditions ; If multiple variables are abnormal at the same time, it is necessary to analyze the global abnormal pattern to determine whether it is a systematic error.

6. The method for predicting coal seam permeability enhancement effect based on multi-source data fusion according to claim 4, characterized in that: The missing data supplementation process specifically includes: A hybrid prediction method based on time dependency and spatial correlation is used to establish the correlation between time series and spatial relationships and data, and to construct a spatiotemporal-multivariable joint interpolation model. ; The spatiotemporal-multivariate joint interpolation model includes long short-term memory networks and recurrent neural networks. Long short-term memory networks are used to analyze the long-term temporal dependencies of different variables, and recurrent neural networks are used to analyze the spatial correlations of different variables. At the same time, a dynamic compensation mechanism is set up, and confidence calculation is first performed. For each predicted value, the corresponding prediction error is calculated; If the previous anomaly detection indicates that the reliability of the variable data is below the threshold, its confidence is adjusted during completion, and then multi-step forecast smoothing is performed.

7. The method for predicting coal seam permeability enhancement effect based on multi-source data fusion according to claim 1, characterized in that: In the multi-parameter joint prediction model for coal seam permeability enhancement, the multi-granularity time feature extraction network includes a fully connected layer, a convolutional layer, a gated unit and an LSTM layer; the multi-granularity time feature extraction network is used to capture the fused time series representation The short-term time-dependent characteristics implied in , mid-term time-dependent characteristics and long-term time-dependent characteristics ; The fully connected layer is responsible for extracting The convolutional layer, gated unit and LSTM layer respectively extract the primary features to obtain the deep features , and ; Then, the cross-attention mechanism layer is used to capture the correlation between multi-granular features to obtain cross-fusion features. ; The hierarchical linear regression layer designs four prediction branches with the same architecture but different parameters. Each branch corresponds to a target variable. For each target variable Construct a hierarchical linear regression prediction branch, the basic form of which is as follows: in is the cross-fusion feature of the input, is the regression coefficient, is the regression bias, is the interval number index; is the interval number, indicating that different regression coefficients are used for fitting in different data ranges; each branch shares the underlying features but has an independent prediction layer; The prediction results of four key parameters are permeability, gas production, pressure change and fracture density; The multi-parameter joint prediction model for coal seam permeability increase adopts outlier constraints and smoothing strategies when predicting. If the predicted value deviates from the historical trend, it will be corrected.

Images